I really want to understand what the context consumption looks like for this. Is it 10k tokens? Is it 100k tokens?